Finetuned from model : meta-llama/Llama-3.2-3B-Instruct
This is one of my first Reasoning model can have an “aha moment” same as DeepSeek’s R1. We've enhanced the entire GRPO process, making it use 80% less VRAM than Hugging Face + FA2. This allows you to reproduce R1-Zero's "aha moment" on just 7GB of VRAM using llama-3.2-3b. Please note, this isn’t fine-tuning DeepSeek’s R1 distilled models or using distilled data from R1 for tuning. This is converting a standard model into a full-fledged reasoning model using GRPO.
